huskies: merge 967

server/src/agents/pty/runner.rs

@@ -33,6 +33,16 @@ use super::types::{ChildKillerGuard, PtyResult, composite_key};
 /// If the agent committed valid work before crashing, the "work survived" check
 /// in `pipeline::advance` detects the committed code and advances the story to
 /// QA instead of entering the retry/block path.
+///
+/// ## `eager_record` — watchdog-kill race fix (bug 967)
+///
+/// When `Some((project_root, model))` is passed, the blocking thread calls
+/// `session_store::record_session()` immediately when the `"system"` JSON event
+/// is parsed.  This runs inside the OS blocking thread, which cannot be
+/// cancelled by a tokio task abort.  If the watchdog later kills the PTY child
+/// and aborts the spawned tokio task, the session_id is already persisted and
+/// the respawn's `lookup_session()` returns it (warm start instead of cold).
+/// Pass `None` when session persistence is not needed (e.g. in tests).
 #[allow(clippy::too_many_arguments)]
 pub(in crate::agents) async fn run_agent_pty_streaming(
     story_id: &str,
@@ -48,6 +58,7 @@ pub(in crate::agents) async fn run_agent_pty_streaming(
     child_killers: Arc<Mutex<HashMap<String, Box<dyn ChildKiller + Send + Sync>>>>,
     watcher_tx: broadcast::Sender<WatcherEvent>,
     session_id_to_resume: Option<&str>,
+    eager_record: Option<(std::path::PathBuf, String)>,
 ) -> Result<PtyResult, String> {
     let sid = story_id.to_string();
     let aname = agent_name.to_string();
@@ -74,6 +85,7 @@ pub(in crate::agents) async fn run_agent_pty_streaming(
             &child_killers,
             &watcher_tx,
             resume_sid.as_deref(),
+            eager_record,
         )
     })
     .await
@@ -95,6 +107,7 @@ fn run_agent_pty_blocking(
     child_killers: &Arc<Mutex<HashMap<String, Box<dyn ChildKiller + Send + Sync>>>>,
     watcher_tx: &broadcast::Sender<WatcherEvent>,
     session_id_to_resume: Option<&str>,
+    eager_record: Option<(std::path::PathBuf, String)>,
 ) -> Result<PtyResult, String> {
     let pty_system = native_pty_system();
 
@@ -319,6 +332,15 @@ fn run_agent_pty_blocking(
                     .get("session_id")
                     .and_then(|s| s.as_str())
                     .map(|s| s.to_string());
+                // Eagerly persist the session_id so it survives a watchdog kill
+                // that aborts the tokio task before run_agent_spawn's
+                // record_session() call (bug 967).  Runs in the OS blocking
+                // thread — not cancellable by tokio task abort.
+                if let (Some(sid), Some((root, model))) = (&session_id, &eager_record) {
+                    crate::agents::session_store::record_session(
+                        root, story_id, agent_name, model, sid,
+                    );
+                }
             }
             // With --include-partial-messages, thinking and text arrive
             // incrementally via stream_event → content_block_delta. Handle